To enlighten investors on the impact of these breakthroughs and the opportunities they should create, we began publishing Big Ideas in 2017. This annual research report seeks to highlight

Deep Learning Requires Boundless Computational Power
$0
$1
$100
$10,000
$1,000,000
$100,000,000
$10,000,000,000
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
DQN
AlphaFold 2
GPT-3
GPT-2
NMT
ResNet
AlexNet
Dropout
Seq2Seq
Inception
Neural Arch Search
AlphaGo Zero
AlphaGo
Tesla Autopilot
Meena
BERT
DeepSpeech 2
TI7 Dota 1v1
Xception
Visualize ConvNets
VGG
Co
st
t
o
T
ra
in
W
it
h
2
0
20
H
ar
d
w
ar
e
$100M
$1B+

---

•
[1] CAGR: Compound Annual Growth Rate.
Forecasts are inherently limited and cannot be relied upon. | Source: ARK Investment Management LLC, 2020 based on data sourced from company derived statistics.
9
Deep Learning
Deep Learning Is Creating A Boom In AI Chips
$0
$5
$10
$15
$20
$25
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
Billion, USD
Total AI Chip Market
$5
$22
33% CAGR
$1
50% CAGR
1
•
As AI training cost grows from $1 to $100 
million per project, specialized processors 
such as GPUs or TPUs will account for a 
majority of the incremental growth.
•
ARK estimates that data center spending on AI 
processors will scale more than four-fold 
during the next five years, from $5 billion a 
year today to $22 billion in 2025.
•
The upcoming “deployment phase” for deep 
learning will democratize access to AI, 
benefitting not only large internet companies 
but also every industry in the economy.

---

•
*A “Petaflop-Day” is performing a quadrillion operations per second for a day. 
Forecasts are inherently limited and cannot be relied upon. | For informational purposes only and should not be considered investment advice, or a recommendation to buy, sell or hold any particular security. 
Source: ARK Investment Management LLC, 2020 based on data sourced from Hernandez, Danny, and Tom Brown. “AI and Efficiency.” OpenAI, OpenAI, May 2020, openai.com/blog/ai-and-efficiency/ .
10